Time Does Not Diversify Risk: Samuelson's Uncomfortable Result

Ask any financial adviser whether a 25-year-old should hold more stocks than a 65-year-old and you will get an immediate yes. The reasoning sounds airtight: the young investor has decades to ride out a crash, so the risk "averages out" over time. Stocks are safer in the long run.

Paul Samuelson published a paper in 1969 showing that, under a clean set of standard assumptions, this reasoning is wrong. Your investment horizon should not affect the fraction of your wealth you put in stocks. Not a little. Not at all.

The paper appeared in the same journal issue as Robert Merton's continuous-time treatment of the same problem, which reached the same conclusion by different mathematics. Between them, they demolished one of the most beloved pieces of investment folklore, and the folklore has never fully recovered, nor fully died.

The problem: does a long horizon make risk safer?

The intuition behind time diversification is a genuine statistical fact, stated wrong. Here is the true version: over a longer period, the average annual return of stocks becomes more predictable. The law of large numbers really does grind down the variance of the average return. Over one year, your annualized return might be anywhere from minus 40 percent to plus 50 percent. Over thirty years, the annualized average will very likely be in a much narrower band.

That is all true. And it is the wrong quantity to care about.

You do not consume your average annual return. You consume your terminal wealth. And the uncertainty in your terminal wealth, in dollars, does not shrink with time. It grows. Over thirty years, the range of possible outcomes for the actual pile of money you end up with is enormously wider than it is over one year. The averaging effect on percentages is real and simultaneously irrelevant to how uncertain your final wealth is.

Samuelson spent a good chunk of his career making this point, sometimes with visible irritation, and it is the intuitive heart of the paper.

The key idea via analogy: the coin flip you cannot un-flip

Suppose you are offered a favorable but risky bet, and you can take it once, or you can take it a hundred times in a row with your whole stack riding each time.

The folklore says: take it a hundred times, because the good and bad outcomes will cancel and you will end up near the expected value. But that is not what happens when you compound. Repeating a risky bet does not cancel the risk. It compounds it. A bad early outcome permanently shrinks the base that every subsequent bet operates on. The outcomes do not average out, they multiply out, and the spread of possible ending wealths widens with every repetition.

So having many periods ahead of you does not make a single risky bet safer. Samuelson made this rigorous with backward dynamic programming: start at the last period, work out the optimal decision there, then step back one period at a time, at each stage choosing optimally given that you will behave optimally afterwards.

Under his assumptions, which are:

• returns are independent and identically distributed over time, so a bad year tells you nothing about next year, and
• the investor has constant relative risk aversion, meaning they care about percentage gains and losses regardless of how rich they are,

the recursion collapses to a startlingly simple answer: invest the same fraction of your wealth in stocks in every single period, regardless of your wealth and regardless of how many periods remain.

The horizon does not appear in the answer. Neither does your wealth. It is a constant, and it is the same constant at age 25 and age 65.

Why the intuition fails, precisely

The two assumptions are doing all the work, and understanding which one your intuition is quietly violating is the real education this paper provides.

If returns are truly independent, then a crash today gives you no reason at all to expect a rebound. There is no mean reversion, no "recovery." The market does not owe you anything. A young investor who lives through a crash is simply poorer, and faces the same future distribution of returns as before, just with less money. Nothing has "averaged out."

If your risk aversion is proportional, then you care about losing 30 percent of your wealth exactly as much whether that is 30 percent of 50,000 dollars or 30 percent of 5 million. So being richer, or having more years ahead, does not change how much percentage risk you want to bear.

The conclusion follows inescapably. If you want to argue for a glide path, you must break one of those two assumptions. Samuelson's contribution was to make that argument impossible to dodge.

Why it mattered

• It set the standard against which all lifecycle advice is now measured. Target date funds, glide paths, and "age in bonds" rules of thumb all now have to explain themselves. That is a much healthier state of affairs than the pre-1969 situation of just asserting that time reduces risk.
• It clarified what a good argument for glide paths looks like. The strongest one, developed later, is human capital. A young worker owns a large, relatively safe asset that never appears in their brokerage account: their future earnings. Counting that, their total wealth is already heavily bond-like, so their financial portfolio should tilt toward stocks to balance it. As they age, human capital runs down, and the financial portfolio must become more conservative to keep total wealth balanced. Note that this justifies a glide path for a completely different reason than time diversification. The horizon is not doing the work. The changing composition of total wealth is.
• The other legitimate escape hatch is predictability. If stock returns mean-revert, so that bad years are genuinely followed by better ones, then the independence assumption breaks and horizon effects reappear. There is evidence for some long-horizon mean reversion, though it is statistically contested and not nearly strong enough to support the confident folklore.
• It exemplifies a way of thinking. Take a widely believed claim, write down the exact assumptions under which you could prove it, and discover that under the natural assumptions you can prove the opposite. That is how you find out what a belief is actually resting on.

The honest limitations

• The assumptions are strong and knowingly so. IID returns and constant relative risk aversion are chosen for tractability. The world has neither, exactly.
• No labor income, no housing, no pensions. The model has a lone investor with a pile of financial wealth and nothing else. That omission is precisely where the strongest counterargument lives.
• It says nothing about behavior. Even if the math says horizon should not matter, a real investor who panic-sells at the bottom of a crash has effectively realized the loss permanently. Recommending high equity allocations to people who will not stick with them is bad advice regardless of what the model says.
• Real people do not have constant relative risk aversion. Someone with a fixed retirement date and a fixed spending need is not maximizing a smooth utility of wealth. They have a goal, and goals create horizon effects the model does not contain.

The one-line takeaway

Samuelson proved that if returns are unpredictable from year to year and you care about percentage losses rather than dollar losses, then the optimal fraction of your wealth in stocks is the same at every age and at every level of wealth, which means the popular claim that "time diversifies risk" is not a mathematical fact but an assumption about the world that you have to argue for explicitly.